This week in non-Microsoft AI

AI benchmarking and release cadence create procurement risk for autonomous use cases

Public leaderboards and model timelines remain useful, but they do not substitute for enterprise validation. This week’s most actionable signals for Czech buyers are around benchmark reliability for agentic tasks and the pace at which vendors refresh frontier models.

Study warning: agent benchmarks can be gamed

Codesota’s 2026 leaderboard compiles coding, math, and tool-use results across major models, and it cites Berkeley RDI findings that several popular agent benchmarks can reach near-perfect scores without solving the underlying tasks.

Treat strong “agent/tool-use” leaderboard scores as a screening signal, not as evidence of reliability in IT operations, support automation, or workflow orchestration.
Ask vendors to show evaluation on your own task set (tickets, knowledge base, internal apps) and to document anti-gaming controls and failure-mode handling.
Use the leaderboard to shortlist candidates, but lock procurement gates to measurable business outcomes (accuracy on your workflows, time-to-resolution, and controllable error rates).

Source — Codesota LLM Benchmark Leaderboard benchmarksagentic-ai

Model release timeline shows rapid obsolescence risk

AI Flash Report’s model release timeline tracks major model launches and highlights how quickly vendors iterate across frontier and open models.

Design integrations with portability in mind (gateway, abstraction layer, logging) so you can swap models without rewriting business logic.
Structure contracts and budgets to accommodate frequent model refreshes, re-testing, and potential price/performance shifts within the same year.
Set an internal cadence for re-benchmarking and security review when vendors change model versions, context limits, or tool-use behavior.

Source — AI Flash Report model-releasesvendor-strategy

OpenAI coverage tracker helps governance stay current

AI Weekly’s OpenAI tracker aggregates ongoing coverage across product changes, ecosystem activity, and scrutiny, which can help enterprises monitor a fast-moving supplier landscape.

Use a single tracker to support governance workflows (change tracking, vendor risk review, DPIA inputs) when capabilities and policies move faster than quarterly steering cycles.
Treat aggregated media signals as a prompt to verify primary documentation (terms, data handling, retention, residency) before expanding usage to sensitive data.
Benchmark other vendors against the most widely covered feature set and pricing trends to strengthen negotiation positions and multi-vendor planning.

Source — AI Weekly vendor-monitoringgovernance